What happened so far…

In the two previous posts, I took a closer look at the advantages and disadvantages of machine translation—some of it in great detail! (Again congratulations to all who read the whole second post of this series!)

Now that we know more about the positive and potentially problematic aspects of machine translation and generative AI, I would like to take a closer look at specific use cases for machine translation and cases where a human translation would be better suited.

Use cases

Machine translation

Our modern, connected world offers a large number of scenarios where machine translation is not only useful but even necessary. (Even as a trained translator, I am more than happy to admit that.) But what are they?

Getting the gist

Let us say you are looking for information or data for your next project and find a promising website or an interesting paper in a language you do not speak. Within a few clicks and only a couple of seconds, machine translation can tell you whether the source is actually as promising as it seems. A professional human translation would be superfluous in this case.

Text examples: third-party websites, scientific papers, journal articles, etc.

Composing messages

Do you regularly write or receive e-mails and other messages in a language you do not speak well or are not entirely comfortable writing texts in? Or do you know that language quite well but are not 100% sure you are always phrasing things exactly right? A good machine translation tool is worth its weight in gold in this case. Or you could try generative AI: enter a short description of what you would like to say and in which language and you receive a very good draft in just a few seconds! (At least in the “big” languages with close relation to English, such as French, Spanish or German. Translations into rarer languages or ones that are very different to your language might not be quite as good.)

Text examples: e-mail, social media DMs, etc.

Short-lived material

Content is king and creating new content regularly is rewarded by search engines. This means that many texts nowadays are short-lived. This is another case where machine translation comes to the rescue: they tend to be less critical content and due to the quick changes that have to happen, a professional human translation would be too expensive for the short lifetime of the text.

Text examples: customer reviews, etc.

Human translation

Machine translation has numerous uses where human translation is superfluous, but sometimes human expertise is key to success.

Marketing and advertising materials

Marketing and advertising are two areas that almost always require human expertise. Those texts are about directly talking to an audience, engaging them, evoking certain emotions and making people take certain actions. How to best achieve this, however, varies from culture to culture and therefore from language to language. As explored in last month’s post, cultural conventions are not something machine translation can deliver. Direct machine translation therefore often offers much less success than translations by experienced translators and transcreators,* who adapt texts according to those conventions and therefore make them more attractive to the intended audience.

Text examples: websites, brochures, product descriptions (for marketing purposes), etc.

Health and safety

Where texts are crucial for health and safety and dangerous situations could arise from mistranslations, you again need human expertise. Human translators can spot potential ambiguities or discrepancies in texts that could lead to misunderstandings and are often well-versed in regulations in the area they specialise in. Especially in medical contexts, it is advisable to rely on human translations.

Text examples: patient reports, questionnaires, safety instructions, etc.

Cultural references

Many texts contain cultural references, often without us noticing. Or at least until we come across references they do not stem from our culture and that were just translated and not adapted.

What appeals to readers varies from culture to culture and could cause confusion or even deter people from another culture from engaging with a brand or product. It is, again, the domain of human translators to find cultural equivalences for the intended audience.

Text examples: blog posts, texts referring to topics such as sports, TV programmes and similar, TV shows / series / films, etc.

Hybrid scenarios

As is often the case in life, there is always a scale of grey between the extremes, and many translation providers are already providing services in this hybrid area: it is called “post editing” machine translation, meaning that texts are pre-translated with a tool and subsequently checked by humans.

Technical documentation

Technical documents are usually based on formal standards and are therefore a great fit for machine translation. Here, creativity and knowledge of cultural conventions is usually unnecessary or would even be detrimental. But it also contains important information that could affect health and safety or mistakes in use of equipment in case of translation errors. A thorough check by an experienced human post editor is therefore vital.

Text examples: manuals, technical documentation, etc.

Standardised texts

Generally speaking, all texts that are based on a certain standard format that do not require any creativity or cultural knowledge are eligible for machine translation. But where they contain safety-relevant information, they should definitely be checked by human before publication.

Text examples: material safety data sheets, technical product descriptions, etc.

Machine translation or generative AI: who does it better?

With the advent of generative AI some people may wonder: who translates better, machine translation tools or AI?

To keep things short and avoid lengthy technical explanations: I would personally recommend using neural machine translation instead of generative AI for cross-language purposes. The reason for this is, while different prompts can produce interesting results in generative AI tools, they are still prone to hallucinating (“making up things”). Translations-specific engines were, however, built and trained for this purpose and are therefore the recommended option.

Machine translation checklist

No doubt, machine translation has its uses and can speed up the translation process, but should not always be used unsupervised.

Here is a brief checklist to help you decide whether or not to have a text machine translated:

• Does the text contain sensitive/proprietary data or information?
• Does it contain cultural references?
• Do you intend to motivate an audience to engage with your brand or buy your product?
• Are you using this text for brand image purposes?
• Would health and safety be at risk in case of a misunderstanding?

If you answered one or more of the questions above with “yes,” you should definitely approach a human translator. If not, try machine translation, but make sure to have it checked by a native speaker of the language or a professional post editor, just to be on the safe side.

 

This is the end of this short series on machine translation and generative AI, but not quite the end of our exploration of generative AI. With its high-quality and very quick text output, many people seem to wonder whether copywriters are actually becoming increasingly obsolete (just like human translators seem to be in the eye of the general public). But like with human translation we should ask the question: is that actually the case? Come back soon to find out more.

 

* Transcreation is a mix of translation and copywriting. It is especially useful in marketing contexts, where the original text is more of a help for the language professional. They create a new version in the desired language, based on a client brief and the original text.